Multi-Task Contrastive Learning for Skin Lesion Classification and Segmentation
Bibliographic record
Abstract
Accurate skin lesion classification and segmentation from dermoscopic images are essential for the early and effective diagnosis and management of skin cancer. This study aims to develop and evaluate a multi-task convolutional neural network (CNN) that integrates contrastive learning and attention mechanisms for improved feature representations and task-specific performance. We propose the convolution attention (CA) Y-Net, a dual-branch CNN architecture with separate classification and segmentation branches decoding features from a shared encoder. Lesion type is predicted from a global representation that captures salient lesion features, the generation of which in the classification branch is guided by convolutional block attention modules (CBAMs) to enhance focus on the most informative regions. CBAMs are also applied to highlight encoder features, which are then passed to the segmentation decoder via skip connections. The segmentation decoder is further enhanced with deep supervision at multiple decoding stages. To jointly optimize both tasks, we introduce a multi-task contrastive (MuCo) loss, which contrasts lesion and background representations. In addition to pixel-level contrast in the segmentation encoder, MuCo loss introduces a novel cross-branch component that contrasts the global lesion representation from the classification branch with local background representations from the segmentation branch. The model was evaluated on the ISIC 2017 challenge dataset. Both the MuCo loss and CBAMs independently improved classification and segmentation performance, with deep supervision further enhancing segmentation. CA Y-Net achieved the highest mean area under the receiver operating characteristic curve (AUC) for melanoma and seborrheic keratosis classification (92.8%) and the highest Jaccard Index for lesion segmentation (79.6%) among the state-of-the-art methods compared. The high performance afforded by CA Y-Net could potentially enable more accurate diagnosis and quantitative monitoring of skin lesions in clinical applications.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".